Ant Group has open-licensed two trillion-parameter AI models on Hugging Face and ModelScope, giving developers public access to long-context and advanced reasoning systems to accelerate open AI innovation.
Ant Group has open sourced two trillion-parameter frontier AI models — Ling-2.5-1T and Ring-2.5-1T — placing advanced large language and reasoning systems directly in the hands of developers and researchers worldwide. Both models are released under open licences and are publicly downloadable via Hugging Face and ModelScope, reinforcing the company’s push toward an open AI ecosystem.
Ling-2.5-1T serves as the flagship general-purpose model. Built with one trillion parameters, it focuses on higher reasoning efficiency, fine-grained preference alignment and native agent interaction. It supports context windows of up to one million tokens while reducing compute overhead. On AIME 2026, the model matched frontier thinking systems using roughly 5,890 tokens, compared with the typical 15,000–23,000 tokens.
Ring-2.5-1T targets reasoning-heavy workloads and is positioned as the world’s first hybrid linear-architecture thinking model. Designed for advanced logical and mathematical tasks, it achieved gold-tier scores on academic benchmarks, including 35/42 on IMO 2025 and 105/126 on CMO 2025.
Alongside these releases, Ant introduced Ming-Flash-Omni-2.0, a multimodal system that unifies speech, audio and music within a single architecture.
Together, the Ling, Ring and Ming series expand Ant Group’s open-source AI portfolio and signal broader ambitions toward Artificial General Intelligence. The company, also operator of Alipay, aims to make state-of-the-art reasoning and long-context AI more accessible across the global research community.
